Understanding the analysis
How to read ABC, coverage, volatility, and the suggestions that follow validation.
The Analysis step validates the file against the hierarchy and computes a diagnosis of the portfolio. The page refreshes while validation runs. Errors block the forecast; warnings do not.
Summary indicators
| Indicator | Meaning |
|---|---|
| Historical volume | Sum of observed demand |
| Calendar monthly average | Historical volume divided by the calendar months in the dataset |
| DFUs | Number of distinct forecast units |
| Coverage | Share of DFU × month cells that have a value. Below 100% some months or items are missing |
Gaps are months where a DFU has no value. Zeros are months with a recorded demand of zero — the item existed, but demand was nil. Do not treat those as the same thing.
ABC and volatility
ABC classifies series by share of volume, in descending order:
- A — cumulative volume up to 80%
- B — cumulative volume from 80% through 95%
- C — the remainder
The ABC curve and the monthly history stacked by class show where volume is concentrated. A small number of A series often dominate the plan.
Volatility is how much demand swings relative to its average (coefficient of variation). In the ABC × volatility matrix:
- Stable — at or below the median variability of this file;
- Irregular — above that median.
The scatter of average volume × variability (capped at 500 DFUs) uses the same idea: crosses mark the medians.
Completeness and regressors
The completeness block reports calendar months, observations, missing points, and zero observations.
Regressors are split into known future (values on future rows where demand is blank) and historical only. Known-future regressors are what later lets machine-learning models use planned events.
What the diagnosis suggests
The page flags lifecycle patterns and links them to the Parameters step:
| Suggestion | What it means | What to do |
|---|---|---|
| Short-history series | Fewer than 12 monthly observations | Set up cold-start relations |
| New series | Recent activity, none in the prior 3–12 months | Map a source DFU so the forecast does not start from scratch |
| Inactive series | Activity in the prior 3–12 months, none recently | Confirm they should stay in the plan |
| Intermittent series | At least half of observations are zero | Include Croston models; if there are none, Croston is flagged as unnecessary |
Validation report
Download the validation report CSV to inspect errors and warnings line by line. Typical issues include invalid dates, blank dimensions, negative demand, and DFU identifier collisions.
- Review hierarchy returns to the Hierarchy and DFU step without replacing the file.
- Restart upload is the path when the source file itself is wrong.
When the diagnosis looks right, continue to Parameters.